A method for weak fault feature extraction based on variational mode decomposition and stochastic resonance

Abstract

This invention presents a method for weak fault feature extraction based on variational mode decomposition and stochastic resonance, belonging to the field of fault diagnosis for autonomous underwater vehicles (AUVs). The method uses a smoothed pseudo-Wigner distribution to calculate the energy distribution of AUV signals acquired by sensors in the time-frequency domain. The signal is then segmented into energy-concentrated regions and other regions based on the concentration of energy distribution in the time-frequency domain. Variational mode decomposition is then applied to both segments. The intrinsic mode function components obtained after signal decomposition are subjected to bistable stochastic resonance, and the residual signal is subjected to monostable stochastic resonance. The signal after stochastic resonance processing is then reconstructed. Fault features are extracted from the reconstructed signal using a modified Bayesian method, resulting in high fault feature values ​​and a high ratio of fault feature values ​​to noise feature values. This method is applicable to fault feature extraction for AUVs in situations with external ocean current interference and low-level thruster faults.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis for autonomous underwater robots, and in particular to a method for extracting weak fault features based on variational mode decomposition and stochastic resonance. Background Technology

[0002] Autonomous underwater vehicles (AUVs) play a crucial role in marine resource exploration and development, and ensuring their safety is a key research focus in the development and practical application of AUVs. The thruster, as the most critical power component of an AUV, is also one of the main sources of failure. Early-stage thruster failures are often minor, with relatively small power loss and low severity. Early diagnosis of thruster failures can prevent more serious malfunctions.

[0003] Current fault feature extraction methods mostly employ a data-driven approach. A typical method involves first using variational mode decomposition to denoise the signal, and then using modified Bayesian methods for fault feature extraction. This approach performs well for faults with high fault severity. However, it is ineffective for faults with low fault severity, failing to meet the data requirements for subsequent fault severity identification. Summary of the Invention

[0004] To address the aforementioned problems, this invention proposes a weak fault feature extraction method based on variational mode decomposition and stochastic resonance. The method provided by this invention effectively enhances fault feature extraction, especially for cases with low-level thruster faults. It significantly improves the fault feature value and the ratio of fault feature value to noise feature value. When a weak fault occurs in an autonomous underwater vehicle, the method effectively extracts fault features with high fault feature values ​​and a high ratio of fault feature value to noise feature value, which helps improve the accuracy of subsequent fault severity identification and fault detection. Therefore, this invention is significant in the research of fault diagnosis for autonomous underwater vehicles.

[0005] The specific technical solution of this invention: A method for extracting weak fault features based on variational mode decomposition and stochastic resonance, comprising the following steps:

[0006] Step 1: For the signals of the autonomous underwater robot collected by the sensors, the smooth pseudo-Wigner distribution method is used to calculate the energy distribution of the signal in the time and frequency domain;

[0007] Step 2: Based on the results of Step 1, the signal is segmented according to the concentration of energy distribution in the time and frequency domain, and the segmented signals are subjected to variational mode decomposition.

[0008] Step 3: Combining the results of Step 2, perform monostable stochastic resonance on the residual part of the signal obtained after variational mode decomposition, and perform bistable stochastic resonance on the IMF1 part.

[0009] Step 4: Based on the results of Step 3, reconstruct the signal and use the modified Bayesian method to extract fault features from the reconstructed signal.

[0010] As a preferred option: Step 1 specifically involves calculating the smooth pseudo-Wigner distribution of signal X(t) using equation (1):

[0011]

[0012] Where x(t) represents the Hilbert transform of signal X(t), g(u) and h(t) are window functions, and SPWVD x (t,f) represents the energy distribution of the signal in the time-frequency domain;

[0013] According to formula (1), the energy distribution of the autonomous underwater robot signal collected by the sensor in the time and frequency domain is calculated.

[0014] As a preferred option, step (2) specifically involves:

[0015] Based on the energy distribution of the signal in the time-frequency domain obtained in step 1, find the time t corresponding to the point where the energy concentration of the signal is highest in the time-frequency domain. em The signal is segmented into energy concentration regions X. a and the rest of region X b The signal is extracted in 50 frames, 25 frames before and after the moment corresponding to the highest energy point, and then captured as the energy concentration region signal X. a The remaining part of the signal is taken as the signal X of the remaining region. b ;

[0016] The variational mode decomposition method for signals is equation (2):

[0017]

[0018] Among them, {u k}={u1,…,u k} represents the decomposed eigenmode functions, {ω k}={ω1,…,ω k} represents the center frequency corresponding to each intrinsic mode function, and x(t) is the original signal to be decomposed;

[0019] Using the augmented Lagrangian method, the constrained variational problem of equation (2) is transformed into an unconstrained variational problem. The augmented Lagrangian function is introduced as equation (3):

[0020]

[0021] Where α is the penalty factor and λ(t) is the Lagrange multiplier;

[0022] The fish swarm algorithm is used to calculate signal X. a X b The optimal number of modes K and penalty factor α are determined, and the signal is decomposed into a series of IMF components and residuals.

[0023] As a preferred embodiment, step 3 specifically involves:

[0024] The bistable stochastic resonance model is given by equation (4):

[0025]

[0026] in, Let s(t) be the first derivative of the output signal of the bistable system, n(t) be the driving function of the system input, and V′(x) be the noise function of the system input. V′(x) is the first derivative of the bistable state function V(x).

[0027] The bistable state function is equation (5):

[0028] V(x) = -ax 2 / 2+bx 4 / 4 (5)

[0029] Where a and b are the structural parameters of the bistable stochastic resonance system;

[0030] The monostable stochastic resonance model is given by equation (6):

[0031]

[0032] in, Let s(t) be the first derivative of the output signal of the bistable system, s(t) be the driving function of the system input, n(t) be the noise function of the system input, and U′(x) be the first derivative of the monostable state function U(x).

[0033] The monostable state function is given by equation (7):

[0034] U(x) = -ax + bx 4 / 4 (7)

[0035] Where a and b are the structural parameters of the monostable stochastic resonance system;

[0036] Based on equations (4), (5) and the IMF1 signal obtained in step 2, the IMF1 signal components after bistable random resonance are calculated.

[0037] Based on equations (6), (7) and the residual signal obtained in step 2, the residual signal components after monostable stochastic resonance are calculated.

[0038] As a preferred embodiment, step 4 specifically involves:

[0039] The modified Bayesian method is represented by equation (8):

[0040]

[0041] in:

[0042]

[0043]

[0044]

[0045] Where MB(t) represents the extracted fault feature, x(t) represents the signal of the fault feature to be extracted, and N is the time window width. The average value of the fault-free signal. The variance of the fault-free signal

[0046] The residual signal component after monostable random resonance and the IMF1 signal component after bistable random resonance obtained in step 2 are reconstructed, and the fault characteristics of the reconstructed signal are calculated according to equation (8).

[0047] Beneficial effects:

[0048] Existing fault feature extraction methods generally employ variational mode decomposition (MODED) for noise reduction, followed by modified Bayesian methods for fault feature extraction. While this method is effective for faults with high severity, it falls short for low-severity faults, resulting in low ratios of fault feature values ​​to noise levels, failing to meet the data requirements for subsequent fault severity identification. To address this background, this invention proposes a weak fault feature extraction method for autonomous underwater robots based on a combination of variational mode decomposition and stochastic resonance. To enhance the fault feature extraction effect, this method segments the signal according to its energy distribution in the time-frequency domain, dividing it into energy-concentrated regions and other regions. Variational mode decomposition is then performed on both segments. The residuals and IMF1 obtained after decomposition are then subjected to monostable and bistable stochastic resonance, respectively, followed by signal reconstruction. Finally, the reconstructed signal is used for fault feature extraction using modified Bayesian methods. The method provided by this invention effectively enhances fault feature extraction, achieving more accurate fault diagnosis.

[0049] Meanwhile, this method effectively enhances the fault feature extraction effect, especially for cases with low-level thruster faults. It effectively increases the fault feature value and the ratio of fault feature value to noise feature value. When a weak fault occurs in an autonomous underwater robot, the fault features are effectively extracted, and the fault feature value and the ratio of fault feature value to noise feature value are high. This helps to improve the accuracy of subsequent fault degree identification and improve the accuracy of fault detection. Therefore, this invention is of certain significance in the research of fault diagnosis of autonomous underwater robots. Attached Figure Description

[0050] Figure 1 This is a logic diagram of the present invention;

[0051] Figure 2 This is a flowchart of the fault feature extraction method for autonomous underwater robots of the present invention;

[0052] Figure 3 Linear graph comparing the fault characteristics obtained by the method of this invention and the conventional method under the condition that the thruster output loss of this invention is 8%;

[0053] Figure 4 Linear graph comparing fault characteristics obtained by the method of the present invention and the conventional method under the condition that the thruster output loss is 5%. Detailed Implementation

[0054] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0055] In the description of this invention, it should be understood that the terms "upper," "middle," "outer," "inner," etc., which indicate orientation or positional relationship, are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the components or elements referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as limiting this invention.

[0056] Combined with appendix Figure 2 The following describes a method for extracting weak fault features based on variational mode decomposition and stochastic resonance, comprising the following steps:

[0057] Step 1: For the signals of the autonomous underwater robot collected by the sensors, the smooth pseudo-Wigner distribution method is used to calculate the energy distribution of the signal in the time and frequency domain;

[0058] Step 2: Based on the results of Step 1, the signal is segmented according to the concentration of energy distribution in the time and frequency domain, and the segmented signals are subjected to variational mode decomposition.

[0059] Step 3: Combining the results of Step 2, perform monostable stochastic resonance on the residual part of the signal obtained after variational mode decomposition, and perform bistable stochastic resonance on the IMF1 part.

[0060] Step 4: Based on the results of Step 3, reconstruct the signal and use the modified Bayesian method to extract fault features from the reconstructed signal.

[0061] Example 2: Based on Example 1, combined with Appendix Figure 2 To explain, step 1 specifically involves: the method for calculating the smooth pseudo-Wigner distribution of signal X(t) is equation (1):

[0062]

[0063] Where x(t) represents the Hilbert transform of signal X(t), g(u) and h(t) are window functions, and SPWVD x (t,f) represents the energy distribution of the signal in the time-frequency domain;

[0064] According to formula (1), the energy distribution of the autonomous underwater robot signal collected by the sensor in the time and frequency domain is calculated.

[0065] Example 3: Based on Example 2, combined with Appendix Figure 2 To explain, step (2) specifically involves: based on the energy distribution of the signal obtained in step 1 in the time-frequency domain, finding the time t corresponding to the point where the energy concentration of the signal is highest in the time-frequency domain. em The signal is segmented into energy concentration regions X. a and the rest of region X b The signal is extracted in 50 frames, 25 frames before and after the moment corresponding to the highest energy point, and then captured as the energy concentration region signal X. a The remaining part of the signal is taken as the signal X of the remaining region. b ;

[0066] The variational mode decomposition method for signals is equation (2):

[0067]

[0068] Among them, {u k}={u1,…,u k} represents the decomposed eigenmode functions, {ω k}={ω1,…,ω k} represents the center frequency corresponding to each intrinsic mode function, and x(t) is the original signal to be decomposed;

[0069] Using the augmented Lagrangian method, the constrained variational problem of equation (2) is transformed into an unconstrained variational problem. The augmented Lagrangian function is introduced as equation (3):

[0070]

[0071] Where α is the penalty factor and λ(t) is the Lagrange multiplier;

[0072] The fish swarm algorithm is used to calculate signal X. a X b The optimal number of modes K and penalty factor α are determined, and the signal is decomposed into a series of IMF components and residuals.

[0073] Example 4: Based on Example 3, combined with Appendix Figure 2 To explain, step 3 specifically involves:

[0074] The bistable stochastic resonance model is given by equation (4):

[0075]

[0076] in, Let s(t) be the first derivative of the output signal of the bistable system, n(t) be the driving function of the system input, and V′(x) be the noise function of the system input. V′(x) is the first derivative of the bistable state function V(x).

[0077] The bistable state function is equation (5):

[0078] V(x) = -ax 2 / 2+bx 4 / 4 (5)

[0079] Where a and b are the structural parameters of the bistable stochastic resonance system;

[0080] The monostable stochastic resonance model is given by equation (6):

[0081]

[0082] in, Let s(t) be the first derivative of the output signal of the bistable system, s(t) be the driving function of the system input, n(t) be the noise function of the system input, and U′(x) be the first derivative of the monostable state function U(x).

[0083] The monostable state function is given by equation (7):

[0084] U(x) = -ax + bx 4 / 4 (7)

[0085] Where a and b are the structural parameters of the monostable stochastic resonance system;

[0086] Based on equations (4), (5) and the IMF1 signal obtained in step 2, the IMF1 signal components after bistable random resonance are calculated.

[0087] Based on equations (6), (7) and the residual signal obtained in step 2, the residual signal components after monostable stochastic resonance are calculated.

[0088] Example 5, based on Example 4, combined with Appendix Figure 2 To explain, step 4 specifically involves:

[0089] The modified Bayesian method is represented by equation (8):

[0090]

[0091] in:

[0092]

[0093]

[0094]

[0095] Where MB(t) represents the extracted fault feature, x(t) represents the signal of the fault feature to be extracted, and N is the time window width. The average value of the fault-free signal. The variance of the fault-free signal

[0096] The residual signal component after monostable random resonance and the IMF1 signal component after bistable random resonance obtained in step 2 are reconstructed, and the fault characteristics of the reconstructed signal are calculated according to equation (8).

[0097] To verify the effectiveness of the weak fault feature extraction method based on variational mode decomposition and stochastic resonance designed in this invention patent, the following comparative experiment was designed:

[0098] 1) In a simulated autonomous underwater vehicle operating under external ocean current interference, the left thruster experienced an 8% output loss after 100 cycles. The acquired thruster control voltage signal was processed using the traditional variational mode decomposition method, followed by fault feature extraction using a modified Bayesian method. This method was compared with the fault features obtained by the proposed method in this patent, which combines variational mode decomposition and stochastic resonance to process the acquired thruster control voltage signal and then uses a modified Bayesian method for fault feature extraction. The fault features obtained by the two methods were then compared and verified.

[0099] 2) In a simulated autonomous underwater robot operating under external ocean current interference, the left thruster experienced a 5% output loss after 100 cycles. The acquired thruster control voltage signal was processed using the traditional variational mode decomposition method, followed by fault feature extraction using a modified Bayesian method. This method was compared with the fault features obtained by the proposed method in this patent, which combines variational mode decomposition and stochastic resonance to process the acquired thruster control voltage signal and then uses a modified Bayesian method for fault feature extraction. The fault features obtained by the two methods were then compared and verified.

[0100] In the simulation experiment verification process, the same original data were used in both the traditional method and the method designed in this invention, and the verification carrier was the "Beaver II" underwater robot.

[0101] The comparison results obtained by the conventional method and the method designed in this invention are as follows: Figure 3 , 4 As shown.

[0102] from Figure 3 Based on the given thruster fault characteristics, the fault feature value extracted by the traditional method is 0.17, and the ratio of the fault feature value to the noise value is 1.57. The fault feature value extracted by the method of this invention is 0.24, and the ratio of the fault feature value to the noise value is 2.59. Compared with the traditional method, the fault feature value and the ratio of the fault feature value to the noise feature value are improved by 37.65% and 65.24%, respectively, indicating that the method of this invention has a significant effect on fault feature extraction.

[0103] from Figure 4 Based on the given thruster fault characteristics, the fault feature value extracted by the traditional method is 0.21, and the ratio of the fault feature value to the noise value is 2.77. The fault feature value extracted by the method of this invention is 0.23, and the ratio of the fault feature value to the noise value is 3.53. Compared with the traditional method, the fault feature value and the ratio of the fault feature value to the noise feature value are improved by 11.33% and 27.81%, respectively, indicating that the method of this invention has a significant effect on fault feature extraction.

[0104] In summary, this invention studies a method for extracting fault features from the thrusters of autonomous underwater vehicles (AUVs). Under conditions of external ocean current interference and low thruster fault severity, it significantly enhances the effectiveness of fault feature extraction for AUVs. Compared to traditional methods, the extracted fault feature values ​​and the ratio of fault feature values ​​to noise feature values ​​are significantly improved, which is of great significance for fault diagnosis of underwater robots.

[0105] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0106] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for extracting weak fault features of an autonomous underwater robot based on a combination of variational mode decomposition and stochastic resonance, characterized in that, The steps are as follows: Step (1): For the signals of the autonomous underwater robot collected by the sensor, the smooth pseudo-Wigner distribution method is used to calculate the energy distribution of the signal in the time and frequency domain; Step (2): Based on the results of step (1), the signal is segmented according to the concentration of energy distribution in the time and frequency domain, and the segmented signals are subjected to variational mode decomposition respectively. Step (3): Combining the results of step (2), perform monostable stochastic resonance on the residual part of the signal obtained after variational mode decomposition, and perform bistable stochastic resonance on the IMF1 part. Step (4): Based on the results of step (3), the signal is reconstructed, and the fault features are extracted from the reconstructed signal using the modified Bayesian method. The specific steps (2) are as follows: Based on the energy distribution of the signal in the time-frequency domain obtained in step (1), find the time t corresponding to the point where the energy concentration of the signal is highest in the time-frequency domain. em The signal is segmented into energy concentration regions X. a and the rest of region X b Specifically, 25 frames are extracted before and after the moment corresponding to the highest energy point, for a total of 50 frames, as the energy concentration region signal X. a The remaining part of the signal is taken as the signal X of the remaining region. b ; The variational mode decomposition method for signals is equation (2): （2） in, These are the decomposed eigenmode functions. The center frequencies corresponding to each eigenmode function, The original signal to be decomposed; Using the augmented Lagrangian method, the constrained variational problem of equation (2) is transformed into an unconstrained variational problem; the augmented Lagrangian function is introduced as equation (3): （3） in, As a penalty factor, For Lagrange multipliers; The fish swarm algorithm is used to calculate signal X. a X b The most suitable number of modes K and penalty factor The signal is decomposed into a series of IMF components and residuals. The specific steps (3) are as follows: The bistable stochastic resonance model is equation (4): （4） in, Let be the first derivative of the output signal of the bistable system, s(t) be the driving function of the system input, and n(t) be the noise function of the system input. The first derivative of the bistable state function V(x); The bistable state function is equation (5): （5） Where a and b are the structural parameters of the bistable stochastic resonance system; The monostable stochastic resonance model is given by equation (6): （6） in, Let be the first derivative of the output signal of the bistable system, s(t) be the driving function of the system input, and n(t) be the noise function of the system input. It is the first derivative of the monostable state function U(x); The monostable state function is equation (7): （7） Where a and b are the structural parameters of the monostable stochastic resonance system; Based on equations (4), (5) and the IMF1 signal obtained in step (2), the IMF1 signal components after bistable random resonance are calculated. Based on equations (6), (7) and the residual signal obtained in step (2), the residual signal components after monostable random resonance are calculated.

2. The method for extracting weak fault features of autonomous underwater robots based on a combination of variational mode decomposition and stochastic resonance, as described in claim 1, is characterized in that: The specific steps (1) are as follows: Signal The method for calculating the smooth pseudo-Wigner distribution is equation (1): （1） in, Indicates signal Hilbert transform, , For window functions, Indicates signal Energy distribution in the time-frequency domain; According to formula (1), the energy distribution of the autonomous underwater robot signal collected by the sensor in the time and frequency domain is calculated.

3. The method for extracting weak fault features of autonomous underwater robots based on a combination of variational mode decomposition and stochastic resonance, as described in claim 1, is characterized in that: The specific steps (4) are as follows: The modified Bayesian method is represented by equation (8): （8） in: in, For the extracted fault features, The signal from which fault features are to be extracted is N, where N is the width of the time window. The average value of the fault-free signal. The variance of the fault-free signal; The residual signal component after monostable random resonance and the IMF1 signal component after bistable random resonance obtained in step (2) are reconstructed, and the fault characteristics of the reconstructed signal are calculated according to equation (8).

Images